►
From YouTube: Why Running kubelet on your Vacuum Robot is (not) a Good Idea - Christian Simon, Jetstack
Description
Want to view more sessions and keep the conversations going? Join us for KubeCon + CloudNativeCon North America in Seattle, December 11 - 13, 2018 (http://bit.ly/KCCNCNA18) or in Shanghai, November 14-15 (http://bit.ly/kccncchina18).
Why Running kubelet on your Vacuum Robot is (not) a Good Idea - Christian Simon, Jetstack (Any Skill Level)
The Xiamio Mi Vacuum Robot is an affordable bit of kit, yet comes with a powerful SoC and utilises a wide range of sensors. A talk at 34C3 last year showed how to gain root access to the underlying Ubuntu Linux operating system. Based on this groundwork, the talk will explain how the vacuum can be provisioned as a node in a Kubernetes cluster. From then on, well-known Kubernetes primitives can be used to control it: CronJobs periodically schedule drives, and a custom Prometheus exporter is used to track metrics about a vacuum’s life. Using custom controllers and CRDs, extended features of the vacuum can be utilised: requesting raw sensor readings, dumping a map of your home, and allowing the vacuum to drive custom paths.
About Christian
Christian can rely on his experiences from working as a Jack of all trades Linux sysadmin for 10+ years. This hands-on practices is backed by his academic work around security practices. Lately he’s focused on providing modern DevOps environments through the heavy use of containers. As soon as he is away from his laptop he enjoys beers and pork dishes (due to his Bavarian roots), visiting if not organising gigs or music festivals and keeping fit with swimming, hiking, skiing and lately playing field hockey. Previous talks: - Puppetconf 2017: https://www.youtube.com/watch?v=ThbcHUj70EA
Join us for KubeCon + CloudNativeCon in Barcelona May 20 - 23, Shanghai June 24 - 26, and San Diego November 18 - 21! Learn more at https://kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy and all of the other CNCF-hosted projects.
Join us for KubeCon + CloudNativeCon in San Diego November 18 - 21. Learn more at https://bit.ly/2XTN3ho. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy and all of the other CNCF-hosted projects.
A
A
So
what
about
me
I'm
using
kubernetes
for
yeah
the
best
part
of
the
last
three
years,
I've
set
it
up
in
various
sizes
on
different
architectures,
also
on
some
some
arm
kit
before
and
also
you
might
know,
kublai
go
so
I
abandoned
this
data
bit.
So
that's
now
a
bit
legacy
already
and
I
or
my
colleague
James
started
cert
manager.
So
you
should
use
that
definitely
and
I'm
using
vacuum
cleaners
for
more
than
25
years,
and
you
might
not
believe
that.
A
But
that's
me
so
that
was
my
first
vacuum:
cleaner
back
in
yeah,
the
early
90s
I
guess.
So
what
about
he
got?
What
are
we
going
to
look
at
during
this
talk?
So
first
we're
gonna.
Take
a
quick
look
at
the
hardware
capabilities
that
these
rike
walk
vacuum
robots
have
so
they
have
a
quite
powerful
ARM
chip
in
and
a
couple
of
other
chips.
A
Basically,
there
was
a
talk
at
the
34
c3
event,
where
someone
showed
how
to
get
root
access
to
these
vacuums
and
that's
where,
by
basing
a
lot
of
my
knowledge
from
so
they
have
a
pretty
good
documentation.
I
think
it
was
part
of
a
university
project
and
some
other
guy
reverse
engineered
parts
of
the
vacuum
protocol
I'm.
Using
that
a
lot
I
also
reverse
engineer-
reverse
engineered
a
few
commands
myself.
So
there
are
one
or
two
PRS
and
that's
actually
a
Python
command
line
tool
to
control
them.
A
I
also
took
a
look
at
that
that
really
helped
as
well
to
understand
how
the
API
of
the
whole
vacuum
works.
So
it
was
like
I'm
sure,
mid-february
and
I
got
like
an
email
from
the
CN
CF
that
my
talk
had
accepted.
My
vacuum
was
still
on
the
way
from
China,
so
I
didn't
have
one
in
my
hands
yet
so
I
was
thinking
about
yeah.
What
could
I,
possibly
possibly
do
with
this
vacuum?
Yeah
I
had
a
couple
of
ideas.
A
Maybe
it's
a
good
idea
to
run
a
control
plane
on
them,
so
they
have
battery
the
battery
lasts
for
maybe
two
days,
so
it
would
be
quite
h-a
in
terms
of
power,
outage
and
but
I
think,
that's
probably
not
enough
to
fill
a
cube
cons
lot
because
yeah
other
people
run
kubernetes
on
arm
and
yeah.
Maybe
that
Wi-Fi
mesh
network
might
be
interesting,
but
other
than
that
would
be
quite
standard.
A
So
I
was
thinking
about
yeah,
maybe
running
some
actual
workloads
there
or
container
rising
the
vendors
propriety
software
stack,
yeah
I
figured
it
would
be
probably
quite
hard
to
actually
run
containers
there.
So
you
need
C
groups,
you
need
recent
kernels
and
yeah.
So
my
feeling,
without
even
having
had
one
of
those
vacuums,
are
new.
The
kernel
is
quite
old,
so
maybe
I
shouldn't
try
it
myself
and
another
idea
of
us.
Maybe
some
some
kind
of
combat,
seek
and
destroy
the
row
or
make
the
doramas
destroy
each
other
bird
yeah.
A
The
valued
area
is
not
really
too
much
so
I
thought
like
I'm
gonna
work
and
try
to
replace
parts
of
the
software
stack,
especially
the
ones
that
are
like
yeah
in
someone
else's
cloud,
without
your
control
with
yeah
old
written
software
and
maybe
persist
information
into
the
kubernetes
api.
So
looking
at
the
hardware-
so
you
yeah
have
one
of
those
here.
So
it's
quite
light
and
has
yeah
battery
is
a
set
for
around
three
days.
If
you
don't
run
and
actually
make
it
vacuum.
A
So
you
have
a
couple
of
senses:
error
free,
so
that
is
basically
a
view
where
all
the
senses
are.
So
you
have
a
drop
sensor,
so
it
wouldn't
drive
off
a
clip.
You
have
a
speed
meter
to
understand
where
the
vacuum
is
actually
positioned
and
yeah.
Obviously
the
drive
and
yeah
maybe
going
back
a
bit,
and
so
there
exists
two
hardware
revisions
revisions.
This
is
version
one.
A
The
version
one
is
way
cheaper
than
version
to
you
and
it
differed
in
terms
of
firmware
quite
heavily
so
that
all
the
firmware
wasn't
really
aware
of
the
map
in
the
room.
So
you
couldn't
send
it
to
point
XY.
Is
that
so
I'm
going
to
was
necessary
for
that?
But
luckily
in
March
they
released
a
new
firmware
and
I
haven't
started
yet
so
I
didn't
have
to
do
double
work
of
modifying
a
firmware
that
had
this,
which
you
features
in
so
they
weren't
exposed
to
the
user
at
that
time.
A
But
yeah
you
could
use
like
a
bit
more
advanced
pathfinding
yeah
ways.
Also,
the
the
Wi-Fi
is
quite
a
problematic
thing,
so
it's
only
2.4
gigahertz
and
yeah.
We
are
at
the
conference.
Probably
everyone
experienced
a
couple
of
Wi-Fi
drops,
so
that
is
a
bit
tricky
in
the
later
part
in
the
demo.
I
guess
also
there's
a
small
extension
port
for
USB,
so
USB
on-the-go
I
connected
a
five
gigahertz
adapt
which
this
so
that
allows
you
to
also
connect
hard
disks.
If
you
want
to
run
a
file
server
or
anything
like
that,.
A
The
kernel
is
quite
minimal,
so
you
have
to
compile
your
stuff
yourself,
so
I'm
able
to
compile
articles,
but
it
was
a
longer
night
to
figure
everything
out
in
terms
of
CPUs
or
microcontrollers,
so
that
is
a
cortex
a7
arm
there.
So
quad-core
512
Meg's
of
RAM,
so
you
have
decent
RAM,
at
least
to
run
a
couple
of
go
binaries.
So
you
have
a
4
gigabyte
flash
and
it
comes
usually
with
yeah
with
a
couple
of
partitions,
so
you
can
recover
if
you
have
a
failed
firmware
upgrade
so.
A
Speaking
about
the
firmware,
so
the
vacuum
stack
is
based
on
you
boot.
Then
it's
booting,
a
kernel
3.4,
so
quite
old.
It's
some
some
Android
kernel,
so
it
has
a
heavy
modifications
in
there
and
they
run
a
ubuntu
long-term
service.
So
you
can
actually
go
and
ssh
into
a
robot
and
apt-get
install
Emacs
or
whatever
you
might
want
to
have
the
next
thing
or
the
other
open
source
component
is
player
3.10.
So
players
like
a
rope
but
yeah
it
could
control
daemon.
A
So
it's
supports
like
sensors
and
moving
vehicles,
so
yeah
I,
like
or
at
the
beginning,
I
wanted
to
connect
directly
with
this
robot
API
because
seems
like
to
be
the
point
where
all
the
information
is
in,
but
it
was
actually
quite
hard.
So
I
could
somehow
this
player
version
must
have
been
heavily
modified
or
some
binary
protocol
and
I
couldn't
really
speak
to
it.
A
The
kind
of
client
tools
I
have
so
I
had
to
take
a
bit
of
a
different
approach
and
the
different
approach
uses
the
proprietary
vendor
software,
so
yeah,
unless
you
sure,
have
pronounced
that
correctly.
But
the
vendor
Xiaomi
has
like
some
some
kind
of
cloud
protocol
where
you
can
connect
all
sorts
of
IOT
devices
and
yet
that's
one
component.
Then
the
other
component
is
like
managing
the
robot
and
the
communications
between
the
cloud
and
the
the
kind
of
controller
software.
That's
actually
driving
the
robot
through
connection
with
the
player
API.
A
So
I
tried
to
draw
that
a
bit.
Aren't
you
sure
how
accurate
it
is?
But
so
you
have
player
here
sitting
and
connecting
to
all
the
census,
and
you
have
the
robot
controller,
so
the
green
ones
are
the
quad
propriety
components
and
then
basically
the
app
proxy
is
the
interface
for
the
kind
of
Myo
client,
which
is
the
component
and
connects
it
to
the
cloud.
The
cloud
connection
can
happen
via
TCP
and
UDP.
It's
some
a
AES
encryption
on
top
and
it's
more
Datagram
style.
A
So
UDP
you
can
lose
packages,
so
there
are
sequence
numbers
and
there
was
actually
a
quite
good
C
implementation
that
I
took
and
we
have
worked
as
a
base
for
my
goal
line
cloud
client
down
on
the
on
the
server
side.
You
also
have
some
some
kind
of
bypass
of
this
Milo
client,
which
is
used
for
uploading,
the
cleaning
maps
to
the
kind
of
cloud
and
and
they
go
to
through
a
as
free
bucket
and
directly.
So
you
can't
really
intercept
that
easily.
A
So
how
would
that
work
with
the
smartphone,
apps
or
smartphone
app
connected
to
the
cloud?
And
then
just
the
cloud
proxies
it
through
encrypts
it
and
the
commands?
The
smartphone
sense
can
also
be
sent
locally
like
in
if
you're
in
the
Wi-Fi
it
communicates
directly
with
with
the
device.
So
the
the
basic
model
is
like.
The
vacuum
comes
up
with
an
access
point.
A
A
Android
app
or
iPhone
app
speaks,
so
it's
JSON
based
request
response
or
like
drive
to
that
point
or
start
cleaning,
and
then
basically
we
get
a
response,
okay
or
whatever,
and
then
the
kind
of
Club
protocol,
as
I
said,
wraps
the
JSON
protocol
and
this
used
for
for
insecure
channels
or
for
the
public
Internet.
Just
to
make
sure
you
apply
some
encryption
to
that.
So
why
do
I
want
to
liberate
it
from
the
cloud
as
I
call
it?
A
A
Obviously
it
can
be
enabled
like
from
from
the
cloud
side
of
things
also
who
knows
what's
going
to
happen
in
five
years
time,
if
they're
still
going
to
support
the
robot
or
not
so
maybe
I
can
still
spin
up
a
Cuban
at
this
one
point
eight
ten
in
that
time
and
make
it
connect
to
that.
Also
like
that,
the
manuals
don't
contain
anything
how
they
handle
data
and
what
kind
of
guarantees
you
you're
gonna
get,
and
so
rather
take
control
of
the
the
thing
yourself.
So
how
is
the
whole
cabinet
is
coming
into
play?
A
They're,
basically,
I
used,
kubernetes
or
kubernetes
was
the
first
really
large
scale
distributed
system.
I
worked
with
and
yeah
I
understand
it
quite
well
now
and
yeah
I
think
it
can
be
applied
to
a
lot
of
other
problems,
so
4x
for
the
vacuum,
sending
its
status
information
to
their
to
some
central
point,
and
that's
why
I
wanted
to
use
kubernetes
and
use
some
basic
or
extension
concept
of
it
and
yeah.
You
can
just
benefit
from
all
the
kind
of
tooling
around
Kuban
answers.
A
So
if
you
want
to
make
sure
that
only
you
can
control
your
robot
or
you
want
to
give
someone
else,
access
to
certain
sub
commands,
you
might
be
using
our
back
and
create
certain
rules
for
your
vacuum
and
then
make
sure
nobody
else
is
figuring
out.
The
commands
you
can
use
the
ordered
locks
then,
and
use
cube,
ctlt
to
figure
out
how
many
robots
you
have
and
label
them
into
rooms,
and
whatever
else
so
and
also
you
can
monitor
them.
A
So
it's
quite
easy
to
add
the
annotation
to
make
Prometheus
scrape
your
scrape
your
your
actual
robot
and
use
that
somehow
to
alert
you
when
something
goes
wrong.
So
what
did
I
do
so
I
placed
a
component
under
to
the
robot
and
replaced
the
Myo
client
I
called
it
rock
led,
and
this
component
is
then
taking
care
of
the
communications
with
the
app
proxy,
and
it
also
makes
sure
that
it
forwards
everything
to
the
API
server.
A
Ship
CTL
then
uses
the
the
kind
of
normal
way
of
just
connecting
to
the
API
server,
and
then
things
are
taken
from
there
and
you
make
sure
that
the
whole
thing
yeah
it's
executed
on
the
actual
robot
itself.
So
what
is
the
rocket
exactly
doing
so?
I
called
it
a
fake
crew
blasts,
so
it's
basically
doing
similar
things
done
that
at
the
couplet
is
doing,
but
not
running
any
containers
or
anything
like
that.
So
it's
posting
a
note
update
every
couple
of
seconds.
A
The
reason
for
foregoing
with
that
approach
was
for
me
that
I
wanted
to
use
like
higher-level
object,
so
everything
that
uses
paul
potts,
like
in
the
kind
of
background
like,
for
example,
a
cron
job.
What's
the
thing
I
was
thinking
about
our
jobs
in
general,
you,
you
might
want
to
run
a
regular
cleanup
of
a
certain
room
and
you
can
then
make
sure
that
a
pot
gets
created
whenever
you
want
to
clean
your
room
by
using
pots
there.
It's
the
kind
of
object
that
your
fake
rock
let
fake
couplet
is
watching.
A
So
that
is
like
a
spec
for
for
one
of
these
pots,
so
so
basically
I'm
just
using
the
image
name
as
at
the
command,
and
the
first
argument
is
expected
to
be
adjacent,
so
certainly
not
the
most
cleanest
approach,
but
it
works
reasonably
well
and
then
these
commands
they
are
just
forwarded
to
the
a
proxy
edit
with
a
few
kind
of
values
there
and
basically
the
pot
stages
keeps
the
status
of
the
execution.
So
if
you
got
back
and
okay
or
I
failed,
it
marks
the
poorest,
okay
or
fails,
succeeded
or
failed.
A
So
everything
that
I
I've
done
is
especially
written
in
go
and
I'm
pushed
it
just
to
github.
So
if
you
want
to
run
that
yourself,
it's
probably
not
good
enough
documented,
so
it's
just
an
FTP
file
for
now,
and
so
basically
the
way
to
schedule
a
regular
cleaning
would
be,
for
example,
this
crunch
up
here
so
I
hope
large
enough
for
everyone
to
see.
A
A
A
There
we
are,
this
will
create
a
part
at
the
time
and
should
make
the
robot
move.
I
also
wanted
to
investigate
a
bit
more
about
cube,
CTL
xx,
so
I
knew
that
there's
some
some
kind
of
WebSocket
connection
and
the
idea
was,
you
could
see
TLX
akin
to
the
robot
and
have
some
kind
of
remote
control
features
like
up-down
left-right
clean
go
back
to
the
dock.
A
A
Basically,
I
used
Thea
DS
for
posting
that
status
into
the
kubernetes
api.
The
vacuum
cid
contains
like
a
map
where
the
vacuum
actually
is
and
and
reports
its
path
that
it
drove
through
the
room
and
then
also
has
some
status
objects
like
audio
volume
fan
speed,
whatever
I
think
in
the
future,
somebody
could
implement
a
spec
field
as
well
of
spec
struct.
Currently,
I
don't
have
any
value
there,
so
I'm
just
reporting
status
back
there.
A
That's
how
that
looks
like
so
you
can
get
the
battery
level
and
how
much
area
has
been
cleaned
in
in
square
meters
and
yeah.
Just
a
fan
power
that
is
used
for
for
cleaning.
Another
object
is
the
cleaning
itself,
so
every
run
creates
a
cleaning
object.
It's
actually
persisted
on
the
vacuum
itself.
By
default,
so
there's
a
SQLite
database
containing
all
the
maps
and
there
and
yeah
it
has
a
couple
of
more
information
so
like
how
much
area
was
covered.
Did
it
complete?
What
was
the
error
code?
A
So
the
kind
of
maps
thing
was
quite
interesting,
so
I'm
aware
of
at
least
three
formats
under
vacuum
and
that
contained
a
map
of
the
room,
and
so
basically
you
have
one
in
this
archive
format.
You
have
one
that
is
the
actual
live
map
and
I
really
care
to
get
like
the
life
status.
Art
so
I
had
to
implement
them
both
and
also
the
path
information
you
have
a
couple
of
coordinate
systems
to
convert
and
from
so
I
had
to
use
at
least
free
to
get
like
from
the
vacuum
to
the
browser.
A
In
the
end,
the
whole
thing
I
wasn't
too
sure
if
I
really
should
put
like
the
kind
of
point
information
where
the
room
is
into
the
into
a
llamó
format,
so
I
I
just
embed
a
PNG
as
base64
with
that
information,
because
it's
like
yeah,
a
bit
less
storage
and
every
update
you
have
to
upload
like
the
full
information.
So
you
would
have
a
lot
of
traffic
and
yeah.
That's
why
I'm
just
went
for
the
PNG
and
then
I
also
wanted
to
investigate
a
bit
with
a
front-end
kind
of
side
of
things.
A
I
tried
on
before,
but
I
never
really
used
it
with
the
backend
and
I
really
wanted
to
to
have
like
a
client
girl
speaking
you
speaking
to
the
API
server
and
I'm
updating
a
front-end
with
it
and
so
yeah.
It
actually
worked
quite
nice,
so
I'm
having
an
add-on
front-end
that
draws
the
path
onto
the
maps
that
are
pngs
and
it's
yeah
uses
the
cached
information
in
fly
and
go
to.
A
Yeah
list
objects,
the
ADIZ
and
even
if
something
changed
the
back
and
triggers
the
front
and
hey.
This
is
your
D
with
name
and
namespace
has
changed
so
the
front-end
can
reload
it
and
it
will
just
also
update.
So
that
was
actually
something
that
went
quite
smoothly,
so
I'm
not
too
much
of
a
front-end
a
guy,
but
it
was
really
doable
like
you
know,
let's
say
two
days
or
something
like
that
to
get
something
going
there,
and
so
that's
how
the
thing
looks
like
it's.
A
A
So
that's
our
Airbnb,
which
is
on
a
houseboat
which
was
quite
interesting
as
well,
because
it
leans
a
bit
so
their
robot
is
not
that
safe
there,
and
so
it
marks
areas
where
it
was
actually
quite
difficult
for
him
to
drive.
So
like
this.
This
is
like
the
kitchen
table
and
it
always
got
stuck
here,
so
we
you
can
see
that
in
the
past
there,
so
not
just
like
quickly
about
the
kind
of
metrics
I'm
gonna
export.
A
So
it's
the
same
information
like
in
the
series
I'm
just
exporting,
like
the
total
amount
of
yeah
area
covered
over
this
SQLite
history.
Basically,
so
you
have
a
nice
counter
and
can
draw
it
so
I
didn't
go
that
far
to
run
Prometheus
but
theory.
It
shouldn't
be
a
hard
thing
to
get
those
metrics
recorded,
and
would
you
like
them
later
so
basically,
now
is
the
time
for
the
demo
and
I
have
big
concerns
that
it
might
work
might
not
work.
So
basically
we
tried
it
this
morning.
A
Everything
was
fine,
but
when
I
just
wanted
to
set
up
the
Wi-Fi,
the
robots
didn't
connect.
So
we
can
have
a
look
now
if
it's
better
but
I
would
say
potentially
not
so
that
was
to
drive
this
morning.
So
this
is
this
room.
I
I
never
was
another
room
that
big
with
the
robot,
so
I
wanted
to
have
like
a
couple
of
hours
this
morning,
and
so
basically
that
was
to
drive
that
we
did
this
morning
and
so
I'm
gonna
figure
out
shortly.
If
they're
gonna
be
alive
now
it
doesn't
look
like.
A
So
that
is
a
really
nice
commander,
so
else
so
I
named
the
three
robots
according
to
the
founders
of
kubernetes
I
was
not
more
creative.
Sorry
about
that,
and
so
basically
that
are
the
CRTs
that
I
was
speaking
about,
and
you
can
also
see
the
notes
so
yeah.
The
problem
is,
it
says,
not
ready
yeah,
so
they
should
have
connected.
If
you
would
have
four
si
but
yeah
I'm.
Sorry,
that's
something!
I
can't
show
so
I'm
gonna
run.
A
Yeah,
the
problem
is
I,
can't
easily
go
into
these
robots
now,
because
I
can't
connect
to
the
Wi-Fi,
and
that
was
the
last
thing.
I
tried
before
to
just
set
up
some
Wi-Fi,
so
I'm
hope
hoping
to
get
it
running
for
the
stand
and
the
cube
CTL
exact
thing
I
was
talking
about,
but
yeah
I
won't
be
able
to
show
it
now
here
on
stage
so
I
think
I'm,
suggesting
we're
running
through
the
UI
and
I
can
show
a
few
things
about
the
the
CID
update
mechanism.
A
A
Llamo
missing,
which
actually
contains
like
the
the
information
that
I
clicked
on
so
basically
the
respects
back.
You
can
see
the
argument.
Arguments
are
empty
and
I
wanted
to
have
image
called
start.
A
bit
more
interesting
is
because,
like
you
have
to
convert
a
lot
of
coordinates,
if
you
click
here,
the
front-end
sends
to
the
robot
go
to
this
position,
and
it's
really
really
really
annoyed
that
it's
not
working
because
I.
It's
like
quite
interesting,
because
it's
really
accurate
and
driving.
Exactly
to
that
point.
So
we
should
see
now.
A
So
now,
if
we
look
at
that,
we
should
be
able
to
see
an
image
command
go
to
target,
and
this
are
the
coordinates
that
I
clicked
on
in
the
browser
and
yeah
honored
better
day
the
robot
would
have
driven
to
that
coordinates
and
basically
the
map
gets
live,
updated
every
couple.
Every
time
the
map
changes
I'm,
sending
out
an
update
to
the
kubernetes
api
and
yet
a
changes
there
I
like
immediate.
So
let
me
just
do
like
maybe
a
split
screen
here
so.
A
A
So
it's
only
operating
in
this
smaller
area
and,
for
example,
with
changing
the
fan
power
here,
Elm
or
the
client
go
watches
to
change
this
to
the
events
and
gonna
send
a
notification
to
the
front
end
that
actually
will
change
now
and
once
I
click
enter
the
kind
of
fan
speed
to
the
80
like
right
immediately.
So
that
would
be
something
nice
for
the
kubernetes
dashboard
or
anything
like
that.
A
The
whole
thing
is
like
API
comfortable
with
the
kubernetes
api,
so
the
front-end
can
be
targeted
to
the
kubernetes
api
and
it
will
still
be
able
to
list,
though
so
it
I
definitely
always
thought
about
creating
a
front
end
like
this
for
Co,
DS
and
yeah.
It
was
the
first
time
actually
tried
it.
It's
really
helpful.
A
I
could
easily
see
like
generating
some
some
elm
code
from
the
API
struct
to
be
like
quickly
able
to
decode
it
so
back
to
the
slides
and
a
couple
of
more
ideas,
so
I'm
really
sorry
that
the
demo
didn't
work
out.
So
I
should
have
recorded
a
bit
more
this
morning,
so
yeah.
The
problem
is
yeah
to
do
that
properly
and
run
your
own
vacuum
status,
site
online
and
maybe
sell
it
to
customers.
You
need
to
invest
very
more
so
I
just
I,
just
like
hacked
it
together
in
the
last
two
weeks
and
I.
A
Think
there
are
a
couple
of
problems
like
you
should
use
a
your
own
API
server.
You
should
build
it
from
scratch.
Instead
of
just
using
CC.
You
have
way
more
control.
For
example,
the
list
commands
so
cube,
CG
I'll
get
vacuum
shouldn't
contain
the
path
and
the
maps
because
it
just
makes
it
quite
slow,
so
that
could
be
easily
achievable,
I
guess
for
then
custom,
API,
server
and
yeah.
A
Maybe
I
should
use
a
different
name
space
for
different
tenants.
That
has
problems
with
the
notes
because
notes
on
are
namespace.
Obviously
so
that
leads
to
the
next
problem,
maybe
create
like
a
cron
job
for
a
vacuum
job
and
the
vacuum.
Cron
job
in
CR,
these
or
in
custom
objects.
So
it'll
work
quite
a
lot
so
and
I
just
went
down
the
quick
and
easy
route
and
yeah
as
I
said.
Some
generators
for
Elm
would
be
really
helpful.
I
had
to
every
time
I
changed
it.
A
So
I
want
you
to
write
a
control
that
is
actually
matching
the
maps
of
the
vacuums
and
then
doing
something
more
intelligent
like
on
an
API,
so
have
a
level.
So
that
would
be
an
interesting
controller
yeah.
So
that's
about
it
so
yeah.
The
idea
is
that
at
3
p.m.
somehow
you
can
remote
control
it
using
cue,
CTL
exec.
So
hopefully
the
Wi-Fi
will
work
and
yeah.
That
was
my
talk.
I'm
really
sorry
about
that.
I
couldn't
show
anything
and
yeah.